Research On Garbage Classification Algorithm Based On Deep Learning

Garbage classification is an important link to realize garbage reduction,harmlessness and resource utilization.Traditional garbage sorting is mostly carried out by manual sorting,which has disadvantages such as low sorting efficiency and high labor cost.With the continuous improvement of the level of intelligent equipment in my country,it is possible to use computer vision intelligent equipment for garbage sorting.However,traditional image classification algorithms use manual feature extraction and classification.When the types and amounts of garbage increase,the classification accuracy and efficiency of intelligent equipment will decrease.The successful application of deep learning technology in the field of computer vision has improved the accuracy and efficiency of image classification,making it a trend for deep learning technology to replace traditional image classification algorithms for garbage classification.Therefore,the use of deep learning technology to automatically classify garbage is of great significance.First,this article constructs a garbage classification data set.The source of the data set is various junk pictures that are common on the Internet,and junk pictures that do not meet the requirements are eliminated.The data set uses some junk pictures in the Huawei junk data set,and selects the junk picture data that meets the requirements.The data set collects common garbage pictures in life through camera equipment.This article divides garbage images into four categories: dry garbage,wet garbage,hazardous garbage and recyclables.Secondly,this article constructs a garbage classification model.This article compares the advantages and disadvantages of various classic image classification models,and designs experiments to select an image classification model suitable for this topic.At the same time,the attention mechanism is introduced to improve the model,and the influence of the attention mechanism on the accuracy of the Efficient Net series of networks is studied.This article uses models under different training conditions for ensemble learning to increase the generalization ability of the model and reduce the high variance problem that a single model may produce.The research in this paper finds that the attention mechanism can improve the accuracy of the Efficient Net model,and ensemble learning can improve the accuracy of the ensemble model to a certain extent.Finally,this article uses knowledge distillation technology to teach the knowledge learned from the integrated model to the less complex MobileNet V2 model.Experimental results show that using different knowledge distillation parameters generally improves the accuracy of the MobileNet V2 model.Among them,when the temperature T is 20 and α is 0.7,the accuracy ofthe MobileNet model has the highest improvement,with an improvement accuracy of 1.271%,and the final accuracy of the Mobil Net V2 model reaches 93.494%.The research results of this paper are helpful to understand the influence of garbage classification model and attention mechanism,integrated learning and knowledge distillation on garbage image classification model,and provide theoretical guidance for the practical application of garbage classification.